Twitter, renamed X in 2023, is a microblogging service launched in 2006 on which users post short public messages, reply, repost and follow one another. Its open follower graph and, until 2023, its research API made it one of the most studied platforms in social media research. Website ↗
136 datasets tagged as the platform “Twitter (X)”.
A YouTube (with cross-platform spread on Twitter and Facebook) dataset of 7,659 removed YouTube videos with multilingual titles and descriptions, 249,346 associated tweets, and 18,897,277 Facebook shares, collected by Knuutila (2021) via YouTube Data API and CrowdTangle, tracking videos shared on public social media that were subsequently removed by YouTube for COVID-19 misinformation. Each record contains video titles and descriptions in multiple languages, view counts and subscriber counts where recoverable, Facebook reactions/comments/shares, Twitter and Facebook post ID lists for social spread tracking, removal and publication timestamps, and Wayback Machine archive URLs. The dataset is notable for its one of the few public datasets documenting the complete lifecycle of platform-removed health misinformation, linking video metadata to cross-platform social spread and enabling research into both content and dissemination patterns of COVID-19 misinformation. Intended for research into health misinformation detection, content moderation research, cross-platform information spread analysis, COVID-19 infodemic studies, multilingual misinformation narrative analysis, and platform policy impact assessment.
Twitter conversations around the Ottawa Parliament Hill shooting (October 2014), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 890 threads, each a source tweet and its replies, of which 470 were judged rumours by journalists and 420 not. Every rumour also carries its veracity as later established: 329 true, 72 false and 69 unverified. 12,284 tweets in all, with full tweet JSON. ECHO builds the reply network (11,394 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
A Twitter dataset of 11,929 labeled financial tweets, collected by zeroshot via Twitter API. Each record contains tweet text, sentiment label (positive/negative/neutral), sentiment value (0/1/2), train/valid split. A labeled dataset of tweets about financial news, with sentiment annotations for stock market-related content. Suitable for sentiment analysis, financial NLP, stock market prediction, social media analytics, natural language processing.
A NDS-VU signed network repository dataset of 10,885 users and 251,406 signed interaction edges from Italian referendum Twitter, collected by NDS-VU (Vrije Universiteit Amsterdam, 2016). Each record contains directed signed edges between nodes: sign (+1 or -1) representing positive/negative relationships; no edge weights beyond binary sign. Signed retweet/mention network from the 2016 Italian constitutional referendum on Twitter, where edges represent support (+1) or opposition (-1) interactions between users discussing the referendum. Suitable for political polarization, signed social network analysis, referendum discourse.
A Twitter (OffensEval 2020 — Greek) dataset of 10,287 Greek tweets (2,728 offensive, 7,559 non-offensive), collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), split (train/test). OffensEval 2020 Greek — part of the most popular SemEval task of all time (528 teams), enabling multilingual offensive language research. Suitable for offensive language detection, multilingual NLP, cross-lingual transfer learning, content moderation benchmarking.
A Twitter dataset of 10,033 tweets (0 on-topic, 10,033 off-topic), collected by Olteanu et al. via crowdsourcing for the Queensland Floods 2013 crisis. Each record contains tweet text, on-topic/off-topic label (crowdsourced majority voting), crisis type, country, year. The dataset is notable for its tweets from the Queensland Floods 2013 crisis, labeled by crowdsourcing workers for on-topic/off-topic classification with majority voting from at least 3 workers per tweet. Intended for research into crisis informatics, tweet filtering, on-topic detection, information extraction, crowdsourcing evaluation, event detection, situation awareness.
A Twitter dataset of 10,031 tweets (0 on-topic, 10,031 off-topic), collected by Olteanu et al. via crowdsourcing for the Alberta Floods 2013 crisis. Each record contains tweet text, on-topic/off-topic label (crowdsourced majority voting), crisis type, country, year. The dataset is notable for its tweets from the Alberta Floods 2013 crisis, labeled by crowdsourcing workers for on-topic/off-topic classification with majority voting from at least 3 workers per tweet. Intended for research into crisis informatics, tweet filtering, on-topic detection, information extraction, crowdsourcing evaluation, event detection, situation awareness.
A Twitter dataset of 10,012 tweets (0 on-topic, 10,012 off-topic), collected by Olteanu et al. via crowdsourcing for the Boston Bombings 2013 crisis. Each record contains tweet text, on-topic/off-topic label (crowdsourced majority voting), crisis type, country, year. The dataset is notable for its tweets from the Boston Bombings 2013 crisis, labeled by crowdsourcing workers for on-topic/off-topic classification with majority voting from at least 3 workers per tweet. Intended for research into crisis informatics, tweet filtering, on-topic detection, information extraction, crowdsourcing evaluation, event detection, situation awareness.
A Twitter dataset of 10,008 tweets (0 on-topic, 10,008 off-topic), collected by Olteanu et al. via crowdsourcing for the Sandy Hurricane 2012 crisis. Each record contains tweet text, on-topic/off-topic label (crowdsourced majority voting), crisis type, country, year. The dataset is notable for its tweets from the Sandy Hurricane 2012 crisis, labeled by crowdsourcing workers for on-topic/off-topic classification with majority voting from at least 3 workers per tweet. Intended for research into crisis informatics, tweet filtering, on-topic detection, information extraction, crowdsourcing evaluation, event detection, situation awareness.
A Twitter dataset of 10,006 tweets (0 on-topic, 10,006 off-topic), collected by Olteanu et al. via crowdsourcing for the West Texas Explosion 2013 crisis. Each record contains tweet text, on-topic/off-topic label (crowdsourced majority voting), crisis type, country, year. The dataset is notable for its tweets from the West Texas Explosion 2013 crisis, labeled by crowdsourcing workers for on-topic/off-topic classification with majority voting from at least 3 workers per tweet. Intended for research into crisis informatics, tweet filtering, on-topic detection, information extraction, crowdsourcing evaluation, event detection, situation awareness.
A Twitter dataset of 9,992 tweets (0 on-topic, 9,992 off-topic), collected by Olteanu et al. via crowdsourcing for the Oklahoma Tornado Season 2013 crisis. Each record contains tweet text, on-topic/off-topic label (crowdsourced majority voting), crisis type, country, year. The dataset is notable for its tweets from the Oklahoma Tornado Season 2013 crisis, labeled by crowdsourcing workers for on-topic/off-topic classification with majority voting from at least 3 workers per tweet. Intended for research into crisis informatics, tweet filtering, on-topic detection, information extraction, crowdsourcing evaluation, event detection, situation awareness.
A Twitter (OffensEval 2020 — Arabic) dataset of 9,666 Arabic tweets (1,919 offensive, 7,747 non-offensive), collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), split (train/test). OffensEval 2020 Arabic — part of the most popular SemEval task of all time (528 teams), enabling multilingual offensive language research. Suitable for offensive language detection, multilingual NLP, cross-lingual transfer learning, content moderation benchmarking.
A Twitter dataset of 9,399 tweets from hurricane_irma_2017, collected by Alam et al. for hurricane_irma_2017 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the hurricane_irma_2017 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter (#BlackLivesMatter) dataset of 9,165 tweets (3,084 hate speech, 6,081 non-hate), collected by Giorgi et al. via Twitter API (2020). Each record contains tweet text, binary hate label (0/1), hashtags, source platform, user friends/favourites count, retweet flag. 9,165 BLM-related tweets with hate speech labels — captures online discourse during the 2020 Black Lives Matter movement with 3,084 hate speech examples. Suitable for hate speech detection, social movement analysis, BLM discourse research, Twitter content moderation, political communication, racial justice NLP.
A Twitter dataset of 9,112 tweets from hurricane_harvey_2017, collected by Alam et al. for hurricane_harvey_2017 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the hurricane_harvey_2017 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 7,984 tweets from kerala_floods_2018, collected by Alam et al. for kerala_floods_2018 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the kerala_floods_2018 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 7,613 tweets from hurricane_dorian_2019, collected by Alam et al. for hurricane_dorian_2019 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the hurricane_dorian_2019 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 7,376 tweets from california_wildfires_2018, collected by Alam et al. for california_wildfires_2018 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the california_wildfires_2018 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 7,278 tweets from hurricane_maria_2017, collected by Alam et al. for hurricane_maria_2017 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the hurricane_maria_2017 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 32 nodes, 496 edges, collected by Yang, Ferrara, and Menczer (OSoMe/Indiana University). Each record contains user network (GML), suspicious tweet text, message cascades, targeted media accounts. Network and content evidence of a coordinated inauthentic influence operation targeting the 2020 US presidential election on Twitter.. Suitable for Coordinated inauthentic behavior detection, influence operations research, election integrity analysis..
A Twitter dataset of 6,264 tweets from hurricane_florence_2018, collected by Alam et al. for hurricane_florence_2018 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the hurricane_florence_2018 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 6,184 tweets from Fortune 100 companies, collected by Kevin McElwee via Twitter API. Each record contains tweet text, corporation name, sector, timestamp, hashtags, racial justice label, BLM mention, Juneteenth mention, donation mention. A manually annotated dataset of tweets from Fortune 100 companies during the 2020 Black Lives Matter protests, capturing corporate responses to a major social movement. Suitable for corporate activism, social movement response, political communication, crisis communication, public relations, brand activism.
A Twitter dataset of 5,171 sentiment-labeled tweets about NFL teams, collected by Cameron C (Kaggle user cammyc) via Twitter API. Each record contains tweet text, sentiment (positive/negative/neutral), team, username, timestamp, confidence score. A curated dataset of 3,701 NFL-related tweets with sentiment labels, enabling research on sports fan sentiment, team-specific public opinion, and social media engagement patterns during the NFL season. Suitable for sentiment analysis, sports marketing, fan engagement, public opinion tracking, temporal sentiment analysis.
A Twitter dataset of 4,952 tweets (1,238 rumour, 3,714 non-rumour) from the 2013 Boston Marathon bombings, collected by Zubiaga, Liakata, and Procter — Augmented PHEME (Zenodo, 2019). Each record contains tweet text, binary rumour/non-rumour label, timestamp, reaction thread structure linking source tweets to replies. Augmented extension of PHEME covering the Boston Marathon bombings event with both rumour threads and non-rumour threads, providing a balanced benchmark for rumour detection with reaction graph structure. Suitable for rumour detection, misinformation classification, crisis communication analysis, stance detection, conversational thread modelling.
A Twitter (state-sponsored IO accounts) dataset of 3,836 Russian IRA and 770 Iranian state-sponsored troll accounts with profile metadata, collected by Zannettou et al. via Twitter API (2019). Each record contains hashed user ID, display name, screen name, reported location, profile description, follower/following counts, account creation date, account language. Hashed and anonymized profiles of 4,606 verified state-sponsored troll accounts from Russian IRA and Iranian IO campaigns, enabling detection and characterization of coordinated inauthentic behavior. Suitable for information operations detection, state-sponsored troll identification, coordinated inauthentic behavior, influence campaign analysis, social media manipulation research.
Twitter conversations around the Germanwings plane crash in the French Alps (March 2015), from the PHEME rumour dataset (Zubiaga, Liakata, Procter, Kochkina and colleagues; the complete release for rumour detection and veracity classification, figshare 6392078): 469 threads, each a source tweet and its replies, of which 238 were judged rumours by journalists and 231 not. Every rumour also carries its veracity as later established: 94 true, 111 false and 33 unverified. 4,489 tweets in all, with full tweet JSON. ECHO builds the reply network (4,020 reply links, each reply pointing to the tweet it answers) and labels each tweet with its thread's rumour status and, for source tweets, the veracity. Useful for rumour detection, rumour verification and modelling how conversations unfold. Annotations CC BY 4.0; Twitter retains the rights to tweet content. The archive PHEME_veracity.tar.bz2 is gzip-compressed despite its name (the name is the authors' own, kept so the checksum matches figshare): extract it with `tar xzf`.
A Twitter dataset of 4,465 labeled Twitter accounts (991 bots, 3474 genuine), collected by Ashu273k — curated subset of Cresci et al. (2017). Each record contains user profile features (followers, friends, statuses, favourites, listed count), binary bot/genuine label, user bio text. Small, clean subset of the Cresci-2017 dataset curated for rapid prototyping and benchmarking of bot detection models, with pre-extracted user-level features and binary labels. Suitable for bot detection, social spambot classification, feature selection for account authenticity, rapid prototyping of detection algorithms.
A Twitter dataset of 4,021 tweets referencing 3,611 images by URL, collected by Alam et al. for the Hurricane Irma 2017 crisis. Each record contains tweet text, engagement metrics (retweets, favorites, replies, quotes), user metadata, media attachments, crisis code. The dataset is notable for its multimodal dataset from the Hurricane Irma 2017 crisis, collected from Twitter during the event with full Twitter API metadata. Intended for research into crisis informatics, multimodal learning, disaster response, image classification, tweet filtering, engagement analysis. This copy (the SocialED redistribution) holds the tweet objects only: the images are referenced by URL and the image files are not included, and neither are CrisisMMD's human annotations (informativeness, humanitarian category, damage severity).
A Twitter dataset of 3,995 tweets referencing 4,691 images by URL, collected by Alam et al. for the Hurricane Maria 2017 crisis. Each record contains tweet text, engagement metrics (retweets, favorites, replies, quotes), user metadata, media attachments, crisis code. The dataset is notable for its multimodal dataset from the Hurricane Maria 2017 crisis, collected from Twitter during the event with full Twitter API metadata. Intended for research into crisis informatics, multimodal learning, disaster response, image classification, tweet filtering, engagement analysis. This copy (the SocialED redistribution) holds the tweet objects only: the images are referenced by URL and the image files are not included, and neither are CrisisMMD's human annotations (informativeness, humanitarian category, damage severity).
A Twitter dataset of 3,992 tweets referencing 4,569 images by URL, collected by Alam et al. for the Hurricane Harvey 2017 crisis. Each record contains tweet text, engagement metrics (retweets, favorites, replies, quotes), user metadata, media attachments, crisis code. The dataset is notable for its multimodal dataset from the Hurricane Harvey 2017 crisis, collected from Twitter during the event with full Twitter API metadata. Intended for research into crisis informatics, multimodal learning, disaster response, image classification, tweet filtering, engagement analysis. This copy (the SocialED redistribution) holds the tweet objects only: the images are referenced by URL and the image files are not included, and neither are CrisisMMD's human annotations (informativeness, humanitarian category, damage severity).
A Twitter dataset of 3,933 tweets from cyclone_idai_2019, collected by Alam et al. for cyclone_idai_2019 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the cyclone_idai_2019 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter (OffensEval 2020 — English) dataset of 3,887 English tweets (1,080 offensive, 2,807 non-offensive), collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), split (train/test). OffensEval 2020 English — part of the most popular SemEval task of all time (528 teams), enabling multilingual offensive language research. Suitable for offensive language detection, multilingual NLP, cross-lingual transfer learning, content moderation benchmarking.
A HOT Speech (hate/offensive/toxic speech dataset) dataset of 3,481 social media comments with multi-dimensional annotations, collected by SOMAR / ICPSR via Twitter, YouTube, and Reddit APIs (August 2021). Each record contains comment text, hard and soft labels (hateful, offensive, toxic), composite scores, and 21 researcher codes. multi-platform dataset with rich annotations including both MTurk worker labels and researcher codes. Suitable for hate speech detection, offensive language identification, toxicity classification, and political discourse analysis.
A #petpeeve (annoying behaviors) dataset of 3,375 tweets labeled with annoying behavior categories, collected by Wang and Yang via Twitter API (2015). Each record contains tweet text and annoying behavior category label. The dataset is notable for its dataset for automatic categorization of annoying behaviors using lexical and frame-semantic embeddings. Intended for research into social media analysis, text classification, and behavioral categorization.
A Twitter (OffensEval 2020 — Danish) dataset of 3,290 Danish tweets (425 offensive, 2,865 non-offensive), collected by Zampieri et al. for SemEval-2020 Task 12. Each record contains tweet text, binary offensive label (OFF/NOT), split (train/test). OffensEval 2020 Danish — part of the most popular SemEval task of all time (528 teams), enabling multilingual offensive language research. Suitable for offensive language detection, multilingual NLP, cross-lingual transfer learning, content moderation benchmarking.
A 4chan, Twitter, and Weibo (cross-platform) dataset of 2,455 multimodal posts (1,318 hate, 1,137 non-hate), collected by Mira AI Lab (2024). Each record contains image filename, image OCR text, post text, hate/non-hate binary label, hate category (gender/race/religion/politics/violence/health/etc), and human-annotated reason for classification. Multimodal cross-platform collection hate speech dataset combining image OCR text and post text with fine-grained category labels across 8 hate dimensions — 54% of posts are hate speech with human-annotated reasoning. Suitable for multimodal hate speech detection, cross-platform analysis, hate category classification, image-text hate speech, content moderation research. The package includes the 2,455 meme images (img/) that the records name.
A Twitter dataset of 2,242 tweets from canada_wildfires_2016, collected by Alam et al. for canada_wildfires_2016 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the canada_wildfires_2016 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 2,195 tweets from kaikoura_earthquake_2016, collected by Alam et al. for kaikoura_earthquake_2016 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the kaikoura_earthquake_2016 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 2,015 tweets from puebla_mexico_earthquake_2017, collected by Alam et al. for puebla_mexico_earthquake_2017 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the puebla_mexico_earthquake_2017 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 1,958 tweets from pakistan_earthquake_2019, collected by Alam et al. for pakistan_earthquake_2019 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the pakistan_earthquake_2019 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 1,880 tweets from midwestern_us_floods_2019, collected by Alam et al. for midwestern_us_floods_2019 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the midwestern_us_floods_2019 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 1,654 tweets from hurricane_matthew_2016, collected by Alam et al. for hurricane_matthew_2016 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the hurricane_matthew_2016 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 1,563 tweets from ecuador_earthquake_2016, collected by Alam et al. for ecuador_earthquake_2016 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the ecuador_earthquake_2016 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 1,515 tweets from greece_wildfires_2018, collected by Alam et al. for greece_wildfires_2018 crisis. Each record contains tweet text, humanitarian category label, train/dev/test split. Human-annotated crisis tweets from the greece_wildfires_2018 event. Suitable for crisis informatics, humanitarian response, text classification.
A Twitter dataset of 1,486 tweets referencing 1,590 images by URL, collected by Alam et al. for the California Wildfires 2017 crisis. Each record contains tweet text, engagement metrics (retweets, favorites, replies, quotes), user metadata, media attachments, crisis code. The dataset is notable for its multimodal dataset from the California Wildfires 2017 crisis, collected from Twitter during the event with full Twitter API metadata. Intended for research into crisis informatics, multimodal learning, disaster response, image classification, tweet filtering, engagement analysis. This copy (the SocialED redistribution) holds the tweet objects only: the images are referenced by URL and the image files are not included, and neither are CrisisMMD's human annotations (informativeness, humanitarian category, damage severity).
A Twitter dataset of 1,442 tweets, collected by Olteanu et al. via Twitter API during the 2013 Russia meteor crisis. Each record contains tweet text, crisis type, country, event name. The dataset is notable for its tweets from the 2013 Russia meteor crisis, collected systematically across 26 events for comparative analysis. Intended for research into crisis informatics, disaster response, social media analysis, cross-crisis comparison, emergency communication patterns, situation awareness.
A Twitter dataset of 1,412 tweets, collected by Olteanu et al. via Twitter API during the 2012 Costa Rica earthquake crisis. Each record contains tweet text, crisis type, country, event name. The dataset is notable for its tweets from the 2012 Costa Rica earthquake crisis, collected systematically across 26 events for comparative analysis. Intended for research into crisis informatics, disaster response, social media analysis, cross-crisis comparison, emergency communication patterns, situation awareness.
A Cross-platform meme corpus (4chan, Twitter, Weibo) dataset of 1,318 hate-speech labeled meme records from 3 platforms, collected by Mira AI Lab via cross-platform meme collection. Each record contains image filename, OCR-extracted image text, post caption text, hate speech label, category list (politics/race/religion/etc.), and human-readable annotation reason per record. M3-Hate is the hate-speech filtered subset of the M3 meme dataset, with 1,318 hateful memes across 4chan (1062), Twitter (132), and Weibo (124) — multimodal with both image OCR text and post captions. Suitable for hate speech detection, multimodal meme analysis, cross-platform content moderation, image-text classification, harmful content research. The package includes the 1,318 meme images (img/) alongside the JSON records.
A Twitter dataset of 1,250 tweets, collected by Olteanu et al. via Twitter API during the 2013 Savar building collapse crisis. Each record contains tweet text, crisis type, country, event name. The dataset is notable for its tweets from the 2013 Savar building collapse crisis, collected systematically across 26 events for comparative analysis. Intended for research into crisis informatics, disaster response, social media analysis, cross-crisis comparison, emergency communication patterns, situation awareness.
Facebook
YouTube
Reddit
4chan
Weibo